NVIDIA defines 'AI factories' as a new class of infrastructure that converts energy into tokens — the unit of production for AI reasoning models and autonomous agents. Unlike traditional data centers, AI factories are optimized across the full stack (compute, networking, memory, storage, power, cooling, and software) to maximize tokens per watt and minimize cost per token. The NVIDIA GB300 NVL72 system delivers 50x more tokens per megawatt and 35x lower cost per token compared to the Hopper generation. The upcoming Vera Rubin platform promises further gains. NVIDIA also highlights its Dynamo inference orchestration framework, Omniverse DSX Blueprint for digital twin-based facility planning, and its own internal deployment of hundreds of autonomous AI agents. The piece frames AI factories as the foundational industrial infrastructure of the AI era, analogous to power plants in the industrial age.
